spaCy/spacy/tests/lang/pl/test_text.py

29 lines
581 B
Python
Raw Normal View History

# coding: utf-8
"""Words like numbers are recognized correctly."""
from __future__ import unicode_literals
import pytest
2019-08-20 18:36:34 +03:00
@pytest.mark.parametrize(
"text,match",
[
("10", True),
("1", True),
("10,000", True),
("10,00", True),
("jeden", True),
("dwa", True),
("milion", True),
("pies", False),
(",", False),
("1/2", True),
],
)
def test_lex_attrs_like_number(pl_tokenizer, text, match):
tokens = pl_tokenizer(text)
assert len(tokens) == 1
assert tokens[0].like_num == match